Addressing Complex Matrix Interference Improves Multiplex Food

Article Cite This: Anal. Chem. XXXX, XXX, XXX−XXX

pubs.acs.org/ac

Addressing Complex Matrix Interference Improves Multiplex Food Allergen Detection by Targeted LC−MS/MS Derek Croote,† Ido Braslavsky,†,‡ and Stephen R. Quake*,†,§,∥ †

Department of Bioengineering, Stanford University, Stanford, California 94305, United States Robert H. Smith Faculty of Agriculture, Food, and Environment, The Hebrew University of Jerusalem, Rehovot 7610001, Israel § Department of Applied Physics, Stanford University, Stanford, California 94305, United States ∥ Chan Zuckerberg Biohub, San Francisco, California 94158, United States Downloaded via UNIV AUTONOMA DE COAHUILA on July 24, 2019 at 19:59:36 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

‡

S Supporting Information *

ABSTRACT: The frequent use of precautionary food allergen labeling (PAL) such as “may contain” frustrates allergic individuals who rely on such labeling to determine whether a food is safe to consume. One technique to study whether foods contain allergens is targeted liquid chromatography-tandem mass spectrometry (LC−MS/MS) employing scheduled multiple reaction monitoring (MRM). However, the applicability of a single MRM method to many commercial foods is unknown as complex and heterogeneous interferences derived from the unique composition of each food matrix can hinder quantification of trace amounts of allergen contamination. We developed a freely available, open source software package MAtrix-Dependent Interference Correction (MADIC) to identify interference and applied it with a method targeting 14 allergens. Among 84 unique food products, we found patterns of allergen contamination such as wheat in grains, milk in chocolate-containing products, and soy in breads and corn flours. We also found additional instances of contamination in products with and without PAL as well as highly variable soy content in foods containing only soybean oil and/or soy lecithin. These results demonstrate the feasibility of applying LC−MS/MS to a variety of food products with sensitive detection of multiple allergens in spite of variable matrix interference.

F

allergic individuals will often ignore PAL under the assumption that the risk of allergen contamination in these foods is low and phrasing-dependent.7−9,11 This emergent risk stratification by consumers is potentially dangerous and should be superseded by improved labeling that is more representative of the actual risk of allergen presence. To better understand allergen contamination, sensitive and robust methods are required to quantify allergens in commercial foods. Targeted liquid chromatography−tandem mass spectrometry (LC−MS/MS) is an analytical technique for highly multiplexed analyte detection premised on the enzymatic digestion of proteins into peptides that are temporally separated by liquid chromatography prior to interrogation in a mass spectrometer, often in multiple reaction monitoring (MRM) mode and ideally using stable isotope labeled (SIL) peptide analogs spiked into samples prior to injection (Figure S1A). Numerous recent efforts have expanded the number of allergens quantified in mass

ood allergies are characterized by adverse immunologic reactions to otherwise innocuous food proteins and are estimated to affect 5% of adults and 8% of children.1 The potential severity of a reaction coupled with a current absence of a cure creates a strong societal health imperative for reducing the risk of accidental consumption of an allergen by food-allergic individuals. One important step toward this goal is through commercial food allergen labeling, mandated by legislation such as the 2004 Food Allergen Labeling and Consumer Protection Act in the U.S. or European Directives 2003/89/EC and 2006/142/EC. Numerous countries such as China, Canada, the U.S., as well as the European Union require labeling for products which intentionally contain egg, fish, milk, peanuts, shellfish, soy, tree nuts, and wheat.2,3 Some countries extend this list to include other allergens such as celery, lupin, mustard, sesame, and molluscs. However, these laws exempt raw agricultural commodity commingling during harvest, storage, and transport4 and do not apply to accidental cross-contamination of allergens into unlabeled foods during manufacturing. While allergen awareness and the implementation of allergen control plans by food manufacturers has increased,5,6 so has the use of precautionary allergen labeling (PAL), which confuses allergic consumers.7−10 Parents and © XXXX American Chemical Society

Received: March 18, 2019 Accepted: July 12, 2019

A

DOI: 10.1021/acs.analchem.9b01388 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry spectrometry assays,12−18 but these have largely been limited to one or a few standard matrices and lack a thorough investigation of interference as a potential confounding variable. This is especially worrisome as it is not commonplace yet in the field to use SIL peptides that aid in defining peptide retention times and help mitigate matrix effects. An investigation into interference is critical as food ingredients each have a unique proteomic profile that, when combined into a food and modified through processing such as baking, equate to innumerable potential matrices. Consequently, although MRM transitions are meant to be specific to the peptide of interest, other molecular species in the matrix may unpredictably interfere in transition measurement through coelution and satisfaction of Q1 and Q3 transmission window resolution constraints.19 This challenge is well-known in the field of clinical proteomics19,20 and guidelines for MRM assay development dictate that transitions should be thoroughly screened for matrix-derived interferences.21 However, this ideal approach is not feasible for analyses of food, where each matrix contains a potentially unique interference profile. Here we describe a computational approach for addressing heterogeneous interference derived from diverse food matrices and apply it to understand patterns of allergen contamination in commercial foods with and without PAL.

proteomics of allergen protein extracts (see the Supporting Information). These shotgun proteomics data are available through the ProteomeXchange Consortium via the PRIDE24 partner repository and should serve as a resource for the study of allergens and plant proteomics more broadly. Chosen peptides were then synthesized, and Skyline25 was used to select optimal transitions and establish retention times for synthetic unlabeled peptides and stable isotope labeled (SIL) analogs, the latter of which contain 15N13C-isotopically labeled C-terminal lysine or arginine. We then narrowed the list of candidate peptides with preference for those that lacked cysteine and methionine, ionized well, formed doubly charged precursor ions, belonged to multiple isoforms if an allergenic protein contained such variants, and were distributed along the elution gradient. We also attempted to avoid peptides sharing specificity to other species, but this was not possible in every case (see Supporting Information). The final method contained 265 transitions belonging to 53 peptides (Table S1). For quantification, we used external calibrators with internal standardization by constructing an 8 point calibration curve from 1 to 500 ppm on a food mass basis using a serial dilution of target allergen extracts in a wheat matrix and constant addition of SIL peptides (see the Supporting Information). Allergen concentrations are therefore presented as ppm of allergen, for example soy flour; if desired, ppm of allergenic protein can be calculated using the protein fractions reported in Table S2. Allergen detection limits in the wheat matrix, as measured by each allergen’s most sensitive peptide(s), were 5 ppm for 9 allergens, 10 ppm for 1 allergen, and 25 ppm for 3 allergens (Table S1). The median R2 value of all linear regressions was 0.994. For quantification, the most sensitive peptide “quantifier” was used for each allergen. MADIC Algorithm. MADIC is a Python package for identifying and correcting interference in targeted mass spectrometry data that may arise from complex and heterogeneous sample matrices. MADIC accepts data exported as a Report from open-source software Skyline and thus is compatible with any common instrument vendor whose data can be loaded into Skyline. MADIC can be used from the command line or interactively and contains functionality for preprocessing, quality control, and interference detection and correction. The software, along with a visual tutorial and detailed instructions for installation and use are available on github: https://github.com/dcroote/madic. MADIC implements four quality control filters. First, a transition passes the transition ratio quality control filter if its ratio is within a user-defined tolerance from its reference transition ratio as defined by synthetic standard injections. The tolerance is an adjustable parameter that can be specified in absolute or relative terms compared to the reference. These data have been evaluated with an absolute tolerance of 15%. Second, a transition passes the retention time quality control filter if its retention time is within a user-defined tolerance from the retention time of the SIL peptide transitions. As this tolerance will differ based on chromatographic method and consistency of elution, this tolerance is also an adjustable parameter. These data were evaluated with a tolerance of 4.2 s. Third, a peptide passes a signal-to-noise quality control filter if its peak intensity within retention time boundaries is greater than a user-designed multiple of the median background intensity outside of the retention time boundaries. These data were evaluated with a signal-to-noise threshold of 3. Fourth, a

■

EXPERIMENTAL SECTION/MATERIALS AND METHODS Sample Preparation. Food products were selected predominantly from local grocery outlets with preference toward national brands and those with PAL. For each food, a 100 mg sample encompassing any heterogeneity in food content was homogenized with a 5 mm stainless steel bead using the TissueLyser system (Qiagen) in the presence of 1.6 mL of extraction buffer containing 7 M urea, 50 mM ammonium bicarbonate, and 10 mM dithiothreitol (DTT) for cysteine reduction. Samples were then centrifuged at 20,000g to separate the aqueous, lipid, and insoluble fractions. Protein, present in the aqueous fraction, was quantified using the Pierce 660 nm Protein Assay in microtiter plate format and then alkylated through the addition of iodoacetamide to a final concentration of 20 mM and incubated in the dark for 30 min. Samples were then diluted 1:10 with 50 mM ammonium bicarbonate to reduce urea concentrations to less than 1 M for overnight digestion at 37 °C with Trypsin Gold, mass spectrometry grade (Promega) at a ratio of 1 μg trypsin to 50 μg protein. Pierce LC−MS grade acetonitrile and sequencing-grade trifluoroacetic acid were then added to reach final concentrations of 5% and 0.5%, respectively, in order to quench digestion and retain polar peptides during subsequent desalting with C18 MacroSpin columns or 96-well MacroSpin plates (The Nest Group). Desalted samples were eluted with 120 μL twice and dried via speedvac before being reconstituted and transferred to Waters TruView LCMS Certified Clear Glass 12 × 32 mm Screw Neck Total Recovery autosampler vials such that after SIL peptide addition, targeted LC−MS/MS injections contained a total peptide mass of 20 μg. LC−MS/MS Method Development. For each of the 14 selected allergens (almond, brazil nut, cashew, hazelnut, pistachio, walnut, egg, lupin, milk, mustard, peanut, sesame, soy, and wheat), we selected numerous tryptic peptides belonging to two or more clinically recognized allergenic proteins22 using the Allergen Peptide Browser23 or via shotgun B


Article

Analytical Chemistry

Figure 1. Heterogeneous matrix-specific interference in MRM transitions. (A) Binary heatmap of interference among 147 MRM transitions for common food ingredients rice (sample 58), corn (sample 68), and wheat (sample A10). (B) Example chromatograms illustrating interferences in (A). For comparison, synthetic standards depicting expected transition ratios are included on the right. Peak boundaries defined by SIL peptides are shown as vertical gray lines.

peptide passes the replicate quality control filter if each replicate passes all of the three aforementioned filters. Once quality control is performed, MADIC identifies transitions with interference if they fail the transition ratio quality control filter while also exceeding a transition ratio threshold defined by a scaling function that takes into account the transition’s expected transition ratio. This criterion is necessary to identify interference even within peptides that naturally have ratios skewed toward one high intensity transition. To avoid flagging low intensity noise as interference, a transition’s area must also exceed an area threshold. As instruments differ in background intensity and intensity reporting schemes, these cutoffs are configurable parameters. These data were evaluated with a ratio of 10 and minimum area of 2000.

dominant ingredient. As shown in Figure 2A, all 12 breads made with wheat contained wheat-specific interference in the milk Bos d 9 (α-s1-casein) peptide YLGYLEQLLR, which is often-targeted in LC−MS/MS allergen detection assays.23 Similarly, we found consistent interference in the peanut Ara h 3 peptide SPDIYNPQAGSLK in all seven corn flours tested (Figure 2B). Importantly, the challenge of interference is not unique to our method, instrument, or instrument brand. For example, matrix-specific interference has previously affected allergen detection.26 Moreover, as shown in Figure 2C−E, using our method of interference identification, we found consistent interference in data from an interlaboratory plasma protein MRM assay validation study.27 The fact that this study included transition screening for interference prior to assay deployment emphasizes the challenge of simply avoiding interference even under well-controlled conditions with a single sample matrix. Computational Approach for Addressing MRM Interference. We implemented an algorithm-termed MADIC that identifies interference and enables confident quantification of allergens from targeted LC−MS/MS data acquired using a single instrument method and processed with platform-independent and open source Skyline software.25 Interference detection operates under the assumptions that interference will dominate the transition ratio and greatly exceed the intensity of noise surrounding the SIL-defined peptide retention time window. These aspects are illustrated using the aforementioned milk peptide YLGYLEQLLR in a

■

RESULTS AND DISCUSSION Identifying Matrix-Dependent Transition Interference. Using our scheduled MRM method to quantify food allergen contamination, we encountered heterogeneous and matrix-specific MRM transition interference. For example, as illustrated in Figure 1, common food ingredients rice, corn, and wheat each have unique transition interferences that can inhibit quantification of trace amounts of allergen contamination. Moreover, each instance of interference is unique in peak intensity, peak width, and product ion affected. While interference varied significantly by food product, we observed consistent interference within foods composed of a C


Article


transitions, and a distinguished peak relative to the background. Additionally, the requirement that replicate injections satisfy each of these previous criteria serves as a fourth quality control criterion. Chromatograms shown in Figure 3F−H provide examples of these criteria. This work extends previous efforts limited to transition ratio and replicate filtering28,29 by also implementing signal-to-noise and transition retention time matching criteria and packaging it in freely available, tested, and documented open-source software. The computational approach we developed to address heterogeneous matrix interference offers advantages over alternative strategies for improving assay selectivity. For example, in our hands the occasional reduction in interference using differential mobility spectrometry30 did not outweigh the loss in signal intensity due to decreased transition dwell times (Figure S2). Another approach, MRM3, has successfully been applied to a panel of tree nut allergens,31 but its adoption is challenged by instrumentation requirements, nonspecific secondary product ions, manual scheduling windows, and long dwell times incompatible with a highly multiplexed method containing SIL peptides. Additionally, while highresolution mass spectrometers offer an increased ability to discriminate between target analytes and interfering ions of similar mass in complex matrices, these benefits come at the cost of dynamic range, sensitivity, and measurement variation compared to MRM.32 Our approach, on the other hand, can be implemented without additional hardware or modification to existing MRM data acquisition strategies. Quantifying Allergen Contamination in Commercial Foods. We developed a scheduled MRM method on a triple quadrupole mass spectrometer capable of detecting allergenic proteins belonging to the following 14 allergens: almond, brazil nut, cashew, hazelnut, pistachio, walnut, egg, lupin, milk, mustard, peanut, sesame, soy, and wheat. While we did not generate incurred samples due to the impossibility of reproducing the wide variety of processing observed in commercial foods, we dedicated significant time toward method development. Optimizations aimed at increasing recovery and sensitivity included narrowing the scheduling window duration to 1 min, using Skyline for selection of collision energies and declustering potentials, increasing the chromatographic flow rate to narrow peak width without unduly decreasing points across the peak, modifying the elution gradient length and shape as well as the distribution of peptides along the gradient to maximize transition dwell time (Figure S1B−D), increasing the peptide injection mass (Figure S1E), performing overnight trypsin digestion (Figure S1F), maximizing peptide desalting elution volume (Figure S1G), and optimizing the mass spectrometer cell exit potential (Figure S1H). Quantification of allergens relied on internal standardization using SIL peptides spiked into each sample paired with external calibration based on a serial dilution of allergens in a wheat matrix (see Materials and Methods and Supporting Information). We applied this scheduled MRM method with interference detection to 84 common food products in order to survey patterns of allergen contamination (Figure 4). Most foods belonged to the broad categories of grains, dairy, or snacks. Overall, we observed 23 instances of contamination. Soy was the mostly commonly contaminating allergen, followed by tree nuts, wheat, and milk. Walnut and hazelnut contaminated 1 and 2 products with tree nut PAL, respectively, at concentrations of 5 ppm or less and were the only two tree

Figure 2. Ubiquity of interference. (A) Overlaid chromatograms illustrating consistent YLGYLEQLLR y5 transition interference in 12 breads. (B) Overlaid chromatograms illustrating consistent SPDIYNPQAGSLK y9 transition interference in 7 corn flours. (C−E) Examples of peptides exhibiting reproducible MRM interference across instruments and sites. Data were generated by a large-scale interlaboratory study27 and interferences were identified using MADIC. QC samples in the first column of each row provide a reference for the expected transition ratios. For each set of chromatograms, identified by site and instrument vendor, between 4 and 25 injections have been overlaid after retention time correction.

serial dilution of milk in a wheat matrix. The consistently high y5 transition area originating from interference skews the transition ratio at lower milk concentrations and is overcome only at high milk concentrations (Figure 3A,C−D). Once interference is identified, the peptide can be excluded from analysis, avoiding a false positive result. Alternatively, the affected transition can be selectively excluded, optionally rescuing the ability to confirm presence of the peptide if the remaining transitions pass the quality control metrics discussed below (Figure 3B,E). Importantly, the matrix-specificity of interference and the selective exclusion of transitions in one matrix but not others enables a broadly applicable scheduled MRM method capable of detecting and quantifying allergens in a variety of foods, each with a unique interference profile. Due to the complexity of interference in food, quantification of trace amounts of allergen requires additional quality control measures in addition to interference detection. Three established quantitative aspects of an interference-free chromatogram form the basis of the MADIC implementation: relative transition area ratios similar to synthetic standards, coelution of all individual target transitions with respective SIL D


Article


Figure 3. Transition interference identification and quality control criteria enable high confidence peptide detection. (A) Serial milk dilution in a wheat matrix highlights interference in the y5 transition of milk peptide YLGYLEQLLR. (B) Same as (A) but excluding the y5 transition. (C) Transition areas for chromatograms in (A). (D, E) Transition ratios for chromatograms in (A) and (B), respectively. Expected ratios derived from synthetic standard injections are represented by horizontal lines. (F−H) MADIC avoids false positives with quality control filters. (F) Transitions do not coelute and individual transition peak retention times (arrows) do not match the SIL peptide peak retention time (vertical dotted line). (G) Disagreement in transition retention times and transition ratios between duplicate injections. (H) Insufficient signal-to-noise determined by the ratio of peak apex within the SIL peptide-defined peak boundaries (red line) to median background outside of the peak boundaries (blue line).

food products and geographic locations,33−36 although rates reported for products with PAL may be higher.33,37 PAL was featured on 58 products and individual labels listed anywhere from one to six allergens. The phrasing of PAL had numerous slight variations, but could be grouped into the following three general categories: “may contain,” “same facility,” or “same equipment” (Figure 4). Although the number of products with contamination was small, we did not find that phrasing had any strong relationship with risk of contamination. Studies evaluating the contamination of peanut37 and/or hazelnut,38 with larger numbers of products similarly concluded that PAL phrasing has little bearing on actual risk. Common Patterns of Allergen Contamination. Milk was found at 32 ppm (sample 49) and 20 ppm (sample 33) in two snacks containing chocolate chips. The latter, a chocolate chip cookie, elicited an adverse reaction in a milk-allergic

nuts of six found to contaminate any of the products tested, despite similar or better detection limits for the other tree nuts. In products without PAL, hazelnut was found in two cookies (samples 16, 20) and a basil pesto sauce (sample 17) at less than 5 ppm. Egg was found in one penne pasta product (sample 55) with PAL at 3 ppm. Lupin, a flowering plant in the legume family commonly ground into a gluten-free alternative to wheat flour, was not found in any sample, which is consistent with its infrequent use in the U.S. Mustard similarly did not contaminate any product. Within the products sampled, peanut was the second-most commonly included allergen in PAL, and, despite being detected in all products that listed peanut as an ingredient, was not found in any other products. While this result may relate to our detection limit or use of a roasted peanut flour as a calibrator (Table S1), low peanut contamination rates between 0 and 4.4% have been previously reported in a number of studies across a variety of E


Article


Figure 4. Allergen contamination in 84 unique commercial foods with and without PAL. Samples have been grouped by general category and are labeled with a description and sample ID. Values represent ppm food allergen, for example soy flour, except wheat, for which contamination is illustrated with an “X” due to the use of wheat as the calibration matrix. GF = gluten free. Almond, brazil nut, cashew, hazelnut, pistachio, and walnut have been aggregated as “Tree nuts”. Lupin and mustard have been omitted for visualization purposes as neither contaminated any products.

contaminated breads were we also able to detect soy in the bread crust (Figure 5A). The fact that each pair of crumb and crust originated from the same dough enables this discrepancy to be attributed to temperature- and moisture-dependent effects of baking, which lowered extractable protein (Figure S3A) and reduced peptide intensity in a sample-dependent manner (Figure S3B). Soy was also found to contaminate corn flours. Among the seven corn flours tested, three were found to contain soy, and of these, two contained PAL (Figure 4). Two of the corn flours contained 30 ppm soy or more and a second package of sample 59 contained similar levels of soy contamination (45 ppm, sample 66). We extended our analysis of soy presence to a subset of products that distinctly were labeled as containing soy but where soybean oil and/or the emulsifier soy lecithin were the only soy-contributing ingredients. We found variable amounts of soy in 23 of 30 unique products belonging to this category

individual (private correspondence) and a second, independent package of the same product had nearly identical levels of contamination (23 ppm milk, sample 65). Interestingly, we found no evidence of milk contamination in a fudge cookie (sample 62) from the same manufacturer, which highlights the challenge that allergic consumers face in finding a trustworthy brand. Another pattern we observed was the contamination of grains into other grain-based products. For example, we detected wheat in a corn flour (sample 71), barley flour (sample A8), oat flour (sample A12), and rye flour (sample A15). Soy contaminated grain-based products even more frequently. Soy was found in one gluten-free pasta (sample 53) at 6 ppm as well as in numerous corn flours and breads, some with PAL and some without. Specifically, we found 20 ppm or less of soy in the crumb (inner portion) of two breads with PAL and four breads without PAL. In only two of these F


Article


Figure 5. Soy contamination levels in commercial foods relative to the VITAL reference dose (RD) and ED10 threshold. The 95% CI of the ED10 has been shaded. Conversion from ppm soy for each food was made assuming a 50 g serving size. Pairs of samples marked with an asterisk are separate packages of the same product. (A) Soy contamination in breads and corn flours. Bread A/B designation refers to the crumb (inner) or crust (outer) portion of the bread, respectively. (B) Quantity of soy in 32 products that are labeled as containing soy, but where soy lecithin and/or soybean oil are the only soy-contributing ingredients. Gray bars indicate the presence of only soy lecithin, while black bars indicate the presence of soy lecithin and soybean oil.

exceeded even the estimated ED10 (Figure 5B).49 While these results should be confirmed with larger numbers of products, the practice of ignoring PAL when products contain only soy lecithin and/or soybean oil is potentially dangerous for individuals with a severe soy allergy. For all products, we have only tested the final output of an increasingly complex globalized process involving myriad combinations of sourcing, transport, storage, manufacturing, and packaging steps. Consequently, while it is unknown how or when allergen introduction may have occurred, there are potential explanations for the patterns of contamination we observed. For example, milk often contaminates chocolate,50−52 which may be unsurprising given the difficulty of cleaning shared equipment used to produce both dark and milk chocolate. Cross-contamination of grains, on the other hand, can occur through the standard practice of agricultural comingling, which features shared use of harvesting, transport, and storage equipment. Historical precedent has exempted this type of contamination from labeling requirements, but contamination is common. One study reported soy contamination in 62.8% of wheat flours,4 a striking finding that, when considered with the common use of wheat in bakery products and snacks, may explain our soy lecithin findings as well as findings by the United States Food and Drug Administration in which nearly 25% and 11% of products within these categories were contaminated with soy, respectively.53 Issues with comingling are not restricted to soy; one study found 88% of Canadian oat samples contained above 20 ppm of wheat gluten,54 while others reported dramatically variable contamination frequencies depending on geographical location, food

of food (Figure 5B), which was surprising as allergenic soy protein is unlikely to originate from either soybean oil or soy lecithin. In fact, soybean oil, which is a highly refined oil, is exempt from allergen labeling requirements due to its negligible protein content.39−41 Soy lecithin, on the other hand, has been shown to sometimes contain small amounts of allergenic protein41−43 and, in agreement, we found soy protein only in 1 of 3 soy lecithin products tested (Figure 4). Taking into account the small contribution of soy lecithin by mass to foods, the amount of soy protein contributed by soy lecithin is generally considered insufficient to elicit a reaction in allergic individuals44,45 and some clinicians even suggest that allergic individuals ignore soy lecithin on labels.46 Clinical Reactivity and Allergen Origins. To assess whether the measured soy concentrations in breads, corn flours, and products with soy lecithin and/or soybean oil could be clinically reactive, we first calculated the amount of allergen that would be consumed under the assumption of a 50 g food serving. Next, we compared these amounts with predicted population eliciting dose (ED) values compiled from statistical modeling of oral food challenge data. For example, ED05 represents the dose predicted to elicit a response in 5% of the population. The Voluntary Incidental Trace Allergen Labeling (VITAL) initiative has recommended a reference dose (RD) for most major allergens and for soy is 1 mg based on the lower 95% confidence interval of the ED05.47,48 On the basis of this threshold, the amounts of soy ingested from contaminated breads and corn flours would be unreactive (Figure 5A). For products with soy lecithin and/or soybean oil, on the other hand, seven were above the RD, one of which G


Article

Analytical Chemistry category, and presence of specific labeling such as “glutenfree”.55 While the prevalence of contamination may be concerning, the quantity of allergen must be considered with respect to predicted eliciting doses. For soy, our results and those of the aforementioned studies suggest that few products are predicted to elicit reactions. On the other hand, quantities of wheat gluten exceeding the “gluten-free” 20 ppm threshold have been found frequently in grains but less commonly among products explicitly labeled gluten-free.55−57 The patterns of contamination discussed here are likely a subset of allergen and food category combinations that represent high-risk scenarios. For these scenarios there is an outstanding opportunity for more rigorous testing along the supply chain and for labeling that is more informative than the unregulated, ubiquitous, and unhelpful PAL currently featured on commercial food products.

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. ORCID

Derek Croote: 0000-0003-4907-1865 Ido Braslavsky: 0000-0001-8985-8211

■

Author Contributions

CONCLUSIONS Through a survey of commercial foods, we have highlighted how a single multiplexed scheduled MRM method can overcome complex matrix interference and capture trends in allergen contamination such as milk in chocolate-containing products, soy in breads and flours, wheat in grains, and variable soy content in products with soy lecithin. Although our sample size was small relative to the product market, we found an overall low incidence of contamination and generally low amounts of allergen when contamination was found. To improve communication of actual risk to the allergic consumer, better characterizing scenarios that present high risks of contamination using a modality such as LC−MS/MS should be part of a comprehensive approach involving PAL standardization,58,59 incentives to avoid contamination,60 and the adoption of allergen reference doses by regulatory agencies that would reduce manufacturer uncertainty.10,47,48 Data Availability. Shotgun proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE24 partner repository with the data set identifier PXD007688 (http://dx.doi.org/10.6019/PXD007688). Code and examples for matrix-dependent interference correction are available on Github: https://github.com/dcroote/madic.

■

Liquid chromatography gradient using 0.1% formic acid in water (mobile phase A) and 0.1% formic acid in acetonitrile (mobile phase B); two linear segments were used for peptide elution: the first from 7% B to 21.5% B contained the majority of peptides, while the second from 21.5% B to 30% B contained fewer peptides and was consequently designed with a steeper slope; following peptide elution, the column was washed with 95% B and re-equilibrated (XLSX)

D.C. and I.B. planned and executed experiments and analyzed the data. D.C. developed the software. S.R.Q. directed the research. All authors wrote the manuscript. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS We would like to acknowledge the staff at the Vincent Coates Foundation Mass Spectrometry Laboratory at Stanford University (SUMS): Karolina Krasinska, Allis Chien, and Theresa McLaughlin for helpful discussions related to quantitative method development and instrument performance, as well as Christopher Adams and Ryan Leib for helpful discussions related to shotgun proteomics and data analysis. This research was supported by the Simons Foundation (SFLIFE #288992 to S.R.Q.), a SUMS seed grant, and the Chan Zuckerberg Biohub. D.C. is supported by an NSF Graduate Research Fellowship and the Kou-I Yeh Stanford Graduate Fellowship. I.B. acknowledges support from Stanford University and from The Hebrew University of Jerusalem.

■

REFERENCES

(1) Sicherer, S. H.; Sampson, H. A. J. Allergy Clin. Immunol. 2014, 133, 291−307 quiz 308 . (2) Allen, K. J.; Turner, P. J.; Pawankar, R.; Taylor, S.; Sicherer, S.; Lack, G.; Rosario, N.; Ebisawa, M.; Wong, G.; Mills, E. N. C.; Beyer, K.; Fiocchi, A.; Sampson, H. A. World Allergy Organ. J. 2014, 7, 10. (3) Gendel, S. M. Regul. Toxicol. Pharmacol. 2012, 63, 279−285. (4) Remington, B. C.; Taylor, S. L.; Marx, D. B.; Petersen, B. J.; Baumert, J. L. Food Chem. Toxicol. 2013, 62, 485−491. (5) Gendel, S. M.; Khan, N.; Yajnik, M. J. Food Prot. 2013, 76, 302− 306. (6) Jackson, L. S.; Al-Taher, F. M.; Moorman, M.; DeVries, J. W.; Tippett, R.; Swanson, K. M. J.; Fu, T.-J.; Salter, R.; Dunaif, G.; Estes, S.; Albillos, S.; Gendel, S. M. J. Food Prot. 2008, 71, 445−458. (7) Zurzolo, G. A.; Koplin, J. J.; Mathai, M. L.; Tang, M. K. L.; Allen, K. J. Med. J. Aust. 2013, 198, 621−623. (8) Ben-Shoshan, M.; Sheth, S.; Harrington, D.; Soller, L.; Fragapane, J.; Joseph, L.; St Pierre, Y.; La Vieille, S.; Elliott, S.; Waserman, S.; Alizadehfar, R.; Harada, L.; Allen, M.; Allen, M. H.; Clarke, A. E. J. Allergy Clin. Immunol. 2012, 129, 1401−1404. (9) Noimark, L.; Gardner, J.; Warner, J. O. Pediatr Allergy Immunol 2009, 20, 500−504. (10) Allen, K. J.; Taylor, S. L. J. Allergy Clin Immunol Pract 2018, 6, 400−407. (11) Marchisotto, M. J.; Harada, L.; Blumenstock, J. A.; Bilaver, L. A.; Waserman, S.; Sicherer, S.; Boloh, Y.; Regent, L.; Said, M.;

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.analchem.9b01388. Experimental methods, supporting figures, and references (PDF) (Table S1) Peptide targets for allergen detection with associated detection limits and calibration curve linear regression parameters; (Table S2) Information for each unique sample, including relevant ingredients, precautionary labeling statements, and precautionary phrasing type; NIST = National Institute of Standards and Technology, GF = gluten-free; “A” sample name prefix indicates the allergen was used as a calibrator; for bread samples 37−48, both the crumb (A suffix) and crust (B suffix) were analyzed; not listed: 9, 65, and 66, which are different packages of products 0, 33, and 59, respectively; (Table S3) MRM transitions with associated instrument parameters; and (Table S4) H


Article

Analytical Chemistry Schnadt, S.; Allen, K. J.; Muraro, A.; Taylor, S. L.; Gupta, R. S. Allergy 2016, 71, 1081−1085. (12) Montowska, M.; Fornal, E. LWT - Food Science and Technology 2018, 87, 310−317. (13) Gu, S.; Chen, N.; Zhou, Y.; Zhao, C.; Zhan, L.; Qu, L.; Cao, C.; Han, L.; Deng, X.; Ding, T.; Song, C.; Ding, Y. Food Control 2018, 84, 89−96. (14) Pilolli, R.; De Angelis, E.; Monaci, L. Anal. Bioanal. Chem. 2018, 410, 5653−5662. (15) Parker, C. H.; Khuda, S. E.; Pereira, M.; Ross, M. M.; Fu, T.-J.; Fan, X.; Wu, Y.; Williams, K. M.; DeVries, J.; Pulvermacher, B.; Bedford, B.; Zhang, X.; Jackson, L. S. J. Agric. Food Chem. 2015, 63, 10669−10680. (16) Planque, M.; Arnould, T.; Dieu, M.; Delahaut, P.; Renard, P.; Gillard, N. J. Chromatogr A 2016, 1464, 115−123. (17) Planque, M.; Arnould, T.; Dieu, M.; Delahaut, P.; Renard, P.; Gillard, N. J. Chromatogr A 2017, 1530, 138−151. (18) Gomaa, A.; Boye, J. Food Chem. 2015, 175, 585−592. (19) Gillette, M. A.; Carr, S. A. Nat. Methods 2013, 10, 28−34. (20) Gallien, S.; Duriez, E.; Demeure, K.; Domon, B. J. Proteomics 2013, 81, 148−158. (21) Carr, S. A.; Abbatiello, S. E.; Ackermann, B. L.; Borchers, C.; Domon, B.; Deutsch, E. W.; Grant, R. P.; Hoofnagle, A. N.; Hüttenhain, R.; Koomen, J. M.; Liebler, D. C.; Liu, T.; MacLean, B.; Mani, D. R.; Mansfield, E.; Neubert, H.; Paulovich, A. G.; Reiter, L.; Vitek, O.; Aebersold, R.; Anderson, L.; Bethem, R.; Blonder, J.; Boja, E.; Botelho, J.; Boyne, M.; Bradshaw, R. A.; Burlingame, A. L.; Chan, D.; Keshishian, H.; Kuhn, E.; Kinsinger, C.; Lee, J. S. H.; Lee, S.-W.; Moritz, R.; Oses-Prieto, J.; Rifai, N.; Ritchie, J.; Rodriguez, H.; Srinivas, P. R.; Townsend, R. R.; Van Eyk, J.; Whiteley, G.; Wiita, A.; Weintraub, S. Mol. Cell. Proteomics 2014, 13, 907−917. (22) Radauer, C.; Nandy, A.; Ferreira, F.; Goodman, R. E.; Larsen, J. N.; Lidholm, J.; Pomés, A.; Raulf-Heimsoth, M.; Rozynek, P.; Thomas, W. R.; Breiteneder, H. Allergy 2014, 69, 413−419. (23) Croote, D.; Quake, S. R. Npj Syst. Biol. Appl. 2016, 2, 16022. (24) Vizcaíno, J. A.; Csordas, A.; del-Toro, N.; Dianes, J. A.; Griss, J.; Lavidas, I.; Mayer, G.; Perez-Riverol, Y.; Reisinger, F.; Ternent, T.; Xu, Q.-W.; Wang, R.; Hermjakob, H. Nucleic Acids Res. 2016, 44, D447−56. (25) MacLean, B.; Tomazela, D. M.; Shulman, N.; Chambers, M.; Finney, G. L.; Frewen, B.; Kern, R.; Tabb, D. L.; Liebler, D. C.; MacCoss, M. J. Bioinformatics 2010, 26, 966−968. (26) Planque, M.; Arnould, T.; Delahaut, P.; Renard, P.; Dieu, M.; Gillard, N. Food Chem. 2019, 274, 35−45. (27) Abbatiello, S. E.; Schilling, B.; Mani, D. R.; Zimmerman, L. J.; Hall, S. C.; MacLean, B.; Albertolle, M.; Allen, S.; Burgess, M.; Cusack, M. P.; Gosh, M.; Hedrick, V.; Held, J. M.; Inerowicz, H. D.; Jackson, A.; Keshishian, H.; Kinsinger, C. R.; Lyssand, J.; Makowski, L.; Mesri, M.; Rodriguez, H.; Rudnick, P.; Sadowski, P.; Sedransk, N.; Shaddox, K.; Skates, S. J.; Kuhn, E.; Smith, D.; Whiteaker, J. R.; Whitwell, C.; Zhang, S.; Borchers, C. H.; Fisher, S. J.; Gibson, B. W.; Liebler, D. C.; MacCoss, M. J.; Neubert, T. A.; Paulovich, A. G.; Regnier, F. E.; Tempst, P.; Carr, S. A. Mol. Cell. Proteomics 2015, 14, 2357−2374. (28) Bao, Y.; Waldemarson, S.; Zhang, G.; Wahlander, A.; Ueberheide, B.; Myung, S.; Reed, B.; Molloy, K.; Padovan, J. C.; Eriksson, J.; Neubert, T. A.; Chait, B. T.; Fenyö, D. Methods 2013, 61, 299−303. (29) Abbatiello, S. E.; Mani, D. R.; Keshishian, H.; Carr, S. A. Clin. Chem. 2010, 56, 291−305. (30) Campbell, J. L.; Le Blanc, J. C. Y.; Kibbey, R. G. Bioanalysis 2015, 7, 853−856. (31) Korte, R.; Brockmeyer, J. Anal. Bioanal. Chem. 2016, 408, 7845−7855. (32) Domon, B.; Aebersold, R. Nat. Biotechnol. 2010, 28, 710−721. (33) Ford, L. S.; Taylor, S. L.; Pacenza, R.; Niemann, L. M.; Lambrecht, D. M.; Sicherer, S. H. J. Allergy Clin. Immunol. 2010, 126, 384−385.

(34) Zagon, J.; Dittmer, J.; Elegbede, C. F.; Papadopoulos, A.; Braeuning, A.; Crépet, A.; Lampen, A. J. Food Compos. Anal. 2015, 44, 196−204. (35) Surojanametakul, V.; Khaiprapai, P.; Jithan, P.; Varanyanond, W.; Shoji, M.; Ito, T.; Tamura, H. Food Control 2012, 23, 1−6. (36) Baumgartner, S.; Fürtler-Leitzenberger, I.; Drs, E.; Molinelli, A.; Krska, R.; Immer, U.; Schmitt, K.; Bremer, M.; Haasnoot, W.; Danks, C.; Romkies, V.; Reece, P.; Wilson, P.; Kiening, M.; Weller, M.; Niessner, R.; Corsini, E.; Mendonça, S. Food Contaminants: Mycotoxins and Food Allergens; Siantar, D. P., Trucksess, M. W., Scott, P. M., Herman, E. M., Eds.; ACS Symposium Series; American Chemical Society: Washington, DC, 2008; Vol. 1001, pp 370−381. (37) Hefle, S. L.; Furlong, T. J.; Niemann, L.; Lemon-Mule, H.; Sicherer, S.; Taylor, S. L. J. Allergy Clin. Immunol. 2007, 120, 171− 176. (38) Pele, M.; Brohée, M.; Anklam, E.; Van Hengel, A. J. Food Addit Contam 2007, 24, 1334−1344. (39) Bush, R. K.; Taylor, S. L.; Nordlee, J. A.; Busse, W. W. J. Allergy Clin. Immunol. 1985, 76, 242−245. (40) Awazuhara; Kawai; Baba; Matsui; Komiyama. Clin. Exp. Allergy 1998, 28, 1559−1564. (41) Paschke, A.; Zunker, K.; Wigotzki, M.; Steinhart, H. J. Chromatogr., Biomed. Appl. 2001, 756, 249−254. (42) Porras, O.; Carlsson, B.; Fällström, S. P.; Hanson, L. A. Int. Arch. Allergy Immunol. 2004, 78, 30−32. (43) Müller, U.; Weber, W.; Hoffmann, A.; Franke, S.; Lange, R.; Vieths, S. Zeitschriftfür Lebensmitteluntersuchung und -Forschung A 1998, 207, 341−351. (44) Taylor, S. L.; Remington, B. C.; Panda, R.; Goodman, R. E.; Baumert, J. L. In Handbook of food allergen detection and control; Elsevier, 2015; pp 341−366. (45) Committee on Food Allergies: Global Burden, Causes, Treatment, Prevention, and Public Policy. Finding a path to safety in food allergy: assessment of the global burden, causes, prevention, management, and public policy; Oria, M. P.; Stallings, V. A., Eds.; National Academies Press (US): Washington, DC, 2016. (46) Taylor, S. L.; Baumert, J. L. Allergenicity of Soybean Lecithin: Expert Opinion Statement, Food Allergy Research & Resource Program; University of Nebraska https://farrp.unl.edu/documents/ OpinionsSummaries/2017_0928_Soy%20Lecithin.pdf (accessed Feb 17, 2019). (47) Allen, K. J.; Remington, B. C.; Baumert, J. L.; Crevel, R. W. R.; Houben, G. F.; Brooke-Taylor, S.; Kruizinga, A. G.; Taylor, S. L. J. Allergy Clin. Immunol. 2014, 133, 156−164. (48) Taylor, S. L.; Baumert, J. L.; Kruizinga, A. G.; Remington, B. C.; Crevel, R. W. R.; Brooke-Taylor, S.; Allen, K. J.; Allergen Bureau of Australia & New Zealand; Houben, G. Food Chem. Toxicol. 2014, 63, 9−17. (49) Zhu, J.; Pouillot, R.; Kwegyir-Afful, E. K.; Luccioli, S.; Gendel, S. M. Food Chem. Toxicol. 2015, 80, 92−100. (50) Crotty, M. P.; Taylor, S. L. J. Allergy Clin. Immunol. 2010, 125, 935−937. (51) U.S. Food and Drug Administration. A Survey of Milk in Dark Chocolate Products. https://www.fda.gov/Food/ IngredientsPackagingLabeling/FoodAllergens/ucm446646.htm (accessed Sep 28, 2017). (52) Hefle, S. L.; Lambrecht, D. M. J. Food Prot. 2004, 67, 1933− 1938. (53) Khuda, S. E.; Sharma, G. M.; Gaines, D.; Do, A. B.; Pereira, M.; Chang, M.; Ferguson, M.; Williams, K. M. Food Addit Contam Part A Chem. Anal Control Expo Risk Assess 2016, 33, 1274−1282. (54) Koerner, T. B.; Cléroux, C.; Poirier, C.; Cantin, I.; Alimkulov, A.; Elamparo, H. Food Addit. Contam., Part A 2011, 28, 705−710. (55) Do, A. B.; Khuda, S. E.; Sharma, G. M. J. AOAC Int. 2018, 101, 23−35. (56) Gélinas, P.; McKinnon, C. M.; Mena, M. C.; Méndez, E. Int. J. Food Sci. Technol. 2008, 43, 1245−1252. I


Article

Analytical Chemistry (57) Farage, P.; de Medeiros Nóbrega, Y. K.; Pratesi, R.; Gandolfi, L.; Assunçaõ , P.; Zandonadi, R. P. Public Health Nutr 2017, 20, 413− 416. (58) Hattersley, S.; Ward, R.; Baka, A.; Crevel, R. W. R. Food Chem. Toxicol. 2014, 67, 255−261. (59) Turner, P. J.; Kemp, A. S.; Campbell, D. E. BMJ. 2011, 343, No. d6180. (60) Gupta, R. S.; Taylor, S. L.; Baumert, J. L.; Kao, L. M.; Schuster, E.; Smith, B. M. J. Food Prot. 2017, 80, 1719−1725.

J


Addressing Complex Matrix Interference Improves Multiplex Food

Recommend Documents